WASHINGTON — The number that matters is not the three million pages the Justice Department released under the Epstein Files Transparency Act. It is 19,034.
That is how many distinct actors — individuals, entities, organizations — an open-source software tool called the Epstein Visualizer has identified across 25,232 documents released by the House Oversight Committee. Between those actors, the tool has extracted 107,030 relationships. The result is a navigable map of connections that no congressional investigator, journalist, or member of the public could construct by hand from the raw archive.
The tool, accessible at epsteinvisualizer.com and described in its GitHub repository as the Epstein Document Network Explorer, processes the email corpus using large language models — specifically Anthropic’s Claude — to convert unstructured document text into structured relationship triples: a subject, an action, an object. The pipeline tags each relationship with the source document, making every extracted connection traceable to the original filing.
That traceability is the mechanism that distinguishes it from the wave of secondary aggregators that followed the initial document releases. Clicking on any relationship in the interface opens the underlying document with the relevant entities highlighted. The tool does not assert guilt or culpability; it renders what the documents say, and lets the user follow the thread.
The question it raises is an uncomfortable one for those who assumed the sheer volume of the files would serve as a form of suppression by complexity. A 25,000-document archive is, functionally, inaccessible to most people. The network visualizer turns it into something navigable in a browser window.
The Epstein Files inquiry reached a new threshold last week when Sarah Kellen gave the House Oversight Committee names of two men she said had abused her — the first time a named survivor had gone on record with specific identifications in the congressional process. The testimony arrived as Republican members of the committee moved to cut off Democratic questioning of Attorney General Todd Blanche on three separate occasions during a single hearing, a procedural pattern that drew bipartisan criticism.
Against that political backdrop, the emergence of tools that operate outside the committee room has taken on a specific significance. The Epstein Visualizer is not the only one. A separate project, as public scrutiny of Epstein’s networks has intensified, EpsteinExposed has indexed more than 2.1 million documents, 1,500 persons of interest, and roughly 9,900 emails from DOJ releases, court filings, and FBI disclosures. A researcher named Matt Miller built a parallel interface that plots every email on a time-and-sender grid, turning the correspondence into a shape rather than a stack of files. The Epstein Visualizer’s GitHub repository has accumulated more than 560 stars since its release and continues to ingest newly released documents.
What the network tool cannot resolve is the problem inherent to any AI-assisted extraction at scale: accuracy is probabilistic, not guaranteed. The tool’s own documentation acknowledges that LLM-generated relationship mappings may contain inaccuracies and encourages user cross-verification against the source documents. Some entity names appear in variant forms across thousands of emails; the deduplication pipeline that merges those variants makes judgment calls that a human archivist might make differently. The tool is a reading aid. It is not a verdict.
That caveat has done little to slow its adoption. Since early 2026, the interface has drawn sustained traffic from researchers, journalists, and members of the public seeking to trace specific names through the archive without reading tens of thousands of pages. The timeline view, which sequences relationships chronologically from 1970 to 2025, has been particularly useful for mapping how Epstein’s documented contacts shifted across different periods of his life — the years before his 2008 guilty plea, the period of his supervised release, and the final months before his 2019 arrest.
The broader context is a document landscape that has grown substantially more complex than any single news cycle can contain. The House Oversight Committee’s releases now span more than 33,000 pages from the DOJ alone, with additional material still being processed from the Epstein estate’s production of roughly 95,000 photographs and files. According to ABC News, the DOJ separately withheld nearly three million additional pages, citing the presence of child sexual abuse material and victim privacy obligations. The gap between what has been released and what exists remains, by official count, substantial.
What the Epstein Visualizer offers is not a resolution to that gap. It is a way to work with what is there. Whether the connections it surfaces lead somewhere the committee has not yet followed is a question the tool cannot answer. The investigators who could answer it have not said.
